一种基于局部和全局拟合的混合多相水平集分割模型及算法

Segmentation model and algorithm of a hybrid multiphase level set based on local and global fitting

提出了一种混合多相分割主动轮廓模型,用于精确地分割灰度图像。针对图像去噪和细小边缘的模糊问题,提出多尺度信息增强和各向异性张量扩散结合的方法,并利用快速预分割获得初始轮廓线。在全局项中,用多种局部特征变量重构拉普拉斯能量拟合函数处理大的目标边缘;为了更好地分割拓扑结构复杂的区域,提出了一种分割深度可控的区域分割能量,并构造了计算局部自适应最佳分割阈值的深度系数。在局部项中,采用局部高斯分布拟合处理灰度不均匀区域。模型使用指数函数作为停止速度函数,并加入鲁棒性的曲线演化停止条件,提高了分割效率。新模型适用于分割灰度图中医学核磁共振图像和背景单一的自然图像,通过与最新的多相水平集模型和深度学习模型对照,结果显示新模型能更好地处理目标的拓扑结构,对边界的定位精确,分割精度更高。

In this paper,a hybrid active contour model for multiphase segmentation was proposed to segment grayscale images precisely.To solve the problem of image denoising and the narrow edges blurring during the diffusion process,a method combining multiscale information enhancement and anisotropic tensor diffusion filter was put forward.Then the initial contours were obtained by using rapid pre-segmentation.In the global term,a variety of local feature variables were used to reconstruct the Laplace fitting energy function to segment the large target edges.In order to segment regions with complex topological structure,a regional segmentation model with controllable depth was proposed and the depth coefficient of the adaptive local segmentation threshold was constructed.In the local term,the local gaussian distribution fitting energy was used to process regions with intensity inhomogeneity.The model used the exponential function as the stopping speed function and the robust curve evolution stopping condition was added to it to improve the segmentation efficiency.The model can be applied to the segmentation of medical magnetic resonance images and natural images with simple background.The comparison between the proposed model and the latest multiphase level set model and deep learning model shows that with precise boundary location and higher segmentation precision,the proposed model can better process the topological structure of the target.